OpenAI introduced GPT-6.1 Terra on October 26, a model for professional work combining image analysis, computer-based application use and writing. It is available through the API and in ChatGPT Work and Codex for the Plus, Pro, Enterprise and Edu plans named in the announcement. The company positions it for frequent tasks performed at high volume.
Standard API prices are $1 per million input tokens and $5 per million output tokens. Reading cached input costs $0.05 per million tokens. According to OpenAI, standard prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol. These figures describe the model's published token rates under standard processing, separately from its other available billing options.
Visual analysis and working across applications
GPT-6.1 Terra accepts text and images and produces text. OpenAI describes uses such as matching product photographs to catalog entries, interpreting technical drawings, completing forms and revising documents. Operating applications through their screens is one of the new model's central capabilities, alongside drafting proposals and other professional material for work carried out on a computer.
The API documentation distinguishes these capabilities from connected tools. The model supports computer use, web search, file search, code execution and integrations with services through MCP. Tool calling requires the Responses API. Chat Completions is supported for using the model, while workflows involving tools use the Responses endpoint and the capabilities provided through that interface.
Image generation is available as a connected tool. The model's native output is text, and its native inputs are text and images. Its specification marks audio and video as unsupported modalities. Activities such as editing a video, described in the announcement, concern work through applications and tools, consistent with the computer-use capabilities presented by OpenAI.
One business example mentioned by the company is reading incoming invoices and entering their details into accounting software. Other examples concern technical drawings and documents that must be interpreted before an action in an application. These are uses OpenAI describes for work automation, where analyzing information and interacting with the interface form parts of the same workflow.
Extended context and reasoning levels
The model specification lists a context window of 1,050,000 tokens, with a maximum of 922,000 input tokens and 128,000 output tokens. Its identifier for programmatic access is gpt-6.1-terra. The documented knowledge cutoff is June 1, 2026, separate from information that can be obtained through connected tools when they are available within the application's workflow.
The API supports the reasoning levels none, low, medium, high, xhigh and max, with medium as the default. The model provides streaming responses, structured outputs and function calling. Its documentation presents it for scoped visual analysis, writing and interface-operation tasks within the GPT-6 model family, with the supported tools specified separately from its native input and output modalities.
OpenAI's reported results in professional evaluations
OpenAI evaluated computer use through Agents' Last Exam, a set of professional tasks that includes interpreting architectural drawings and working with specialized applications. At medium reasoning effort, the company reports a score 15 percentage points higher than GPT-5.6 Terra at approximately one-third of the evaluated cost. The comparison concerns that particular set of tasks and testing conditions.
The company also presents results on MMMU-Pro for visual reasoning, where Terra outperforms GPT-5.6 Terra at every evaluated reasoning level. On OSWorld 2.1, which covers multistep tasks through desktop applications, OpenAI reports that the new model at medium effort exceeds the previous Terra generation's best score. These comparisons are part of the company's published evaluation results.
For coding, OpenAI reports 50.3% on Terminal-Bench 4.0 at maximum reasoning effort, compared with GPT-5.6 Terra's published score of 21.5%. The evaluation covers tasks agents solve through the command line. The company continues to position GPT-6.1 Sol and GPT-6 Astra as options for coding work with the highest level of complexity.
In an internal evaluation built around prompts that induce factual errors, OpenAI reports a reduction in error rate from 53.6% to 16.7% at low reasoning effort compared with GPT-5.6 Terra. At maximum effort, the new model's reported rate is 4.3%. These figures describe the company's internal evaluation and the configurations specified for each comparison between the models.
The safety card published alongside the model classifies its cybersecurity capabilities as Critical and its biological and chemical capabilities as High under OpenAI's own evaluation framework. The company states that it uses the safeguards developed for GPT-6.1 Sol and GPT-6 Astra, with monitoring mechanisms adapted to the new model. These are the company's capability classifications.
Charges for caching, long context and processing modes
The cache-read price is separate from the cost of writing to the cache. Under standard processing, cache writes cost $1.25 per million tokens for 30-minute retention and $2.50 for 12-hour retention. These are the published prices before the adjustments applicable to long-context requests. The rates therefore distinguish reused input from the operation that stores it.
For requests exceeding 272,000 input tokens, the documentation specifies double input and cache rates and a 1.5 multiplier for output across the full request. Batch and Flex prices are 50% below Standard. Fast mode costs twice Standard, while regional processing adds 10% where available. These conditions are listed alongside the model's base rates in its technical specification.
API access and use within ChatGPT Work and Codex subscriptions follow their respective product billing models. The announcement specifies availability for Plus, Pro, Enterprise and Edu in Work and Codex, while the API page separately publishes token prices and technical specifications. OpenAI describes Terra as an option for recurring professional work, emphasizing visual analysis and actions across applications.
